The Application Research Of City Meteorological Forecast Based On Genetic Neural Network

Because of neural network method used in meteorological forecast modeling, its initialized weights and threshold is difficult to determine, requiring repeated training to determine the network structure and various parameters, which is easy to cause the over fitting, effecting the network generalization ability. There are many methods of optimize the BP neural network, commonly genetic algorithm is used to optimize neural network weights threshold. However, genetic algorithm has its own shortcoming which it can not overcome by itself. In the initialization group process of genetic algorithm, some individuals'fitness value is too large in the initialization process, and the fitness value tends to centralization at the later stage of genetic algorithm. The paper proposed the fitness value calibration to improve the genetic algorithm. Therefore, before the genetic operators of genetic algorithm fitness value calibration should be carried out, which play the purpose of genetic algorithm optimization . Based on these thoughts above, using the improved genetic algorithm to optimize the weights and threshold value of BP neural network , and then applied to predict Beijing daily maximum and minimum temperatures . The experiment shows that the improved genetic neural network has certain advantages compared with the standard genetic neural network; it improves the predictive power of neural network.